Background: The Boileau classification distinguishes three surgical neck fracture patterns: types A, B, and C. However, the reproducibility of this classification on plain radiographs is unclear. Therefore, we questioned what the interobserver agreement and accuracy of displaced surgical neck fracture patterns is categorized according to the modified Boileau classification. Does the reliability to recognize these fracture patterns differ between orthopedic residents and attending surgeons?
Methods: This interobserver study consisted of a randomly retrieved series of 30 plain radiographs representing clinical practice in a level 1 and a level 2 trauma center. Radiographs were included from patients (≥18 years) who sustained an isolated displaced surgical neck fracture if they were taken ≤1 week after initial injury. A ground truth was established by consensus among three senior orthopedic surgeons. All images were assessed by 17 orthopedic residents and 17 attending orthopedic trauma surgeons.
Results: Agreement for the modified Boileau classification was fair (κ=0.37; 95% confidence interval [CI], 0.36-0.38) with an accuracy of 62% (95% CI, 57%-66%). Comparison of interobserver variability between residents and attending surgeons revealed a significant but clinically irrelevant difference in favor of attending surgeons (0.34 vs. 0.39, respectively, Δ κ=0.05, 95% CI, 0.02-0.07).
Conclusions: The modified Boileau classification yields a low interobserver agreement with an unsatisfactory accuracy in a panel of orthopedic residents and attending surgeons. This supports the hypothesis that surgical neck fractures are challenging to categorize and that this classification should not be used to determine prognosis if only plain radiographs are available.
Keywords: Boileau classification; Interobserver variability; Proximal humerus fracture; Shaft translation; Surgical neck fractures.